Qualcomm Unveils Dragonfly Data Center Chip, Secures Meta Deal and Takes Aim at Nvidia
Miles Bennett
Qualcomm unveiled its data-center chip brand Dragonfly at its June 24 investor day, announced a multi-year CPU deal with Meta, and raised its non-handset revenue target to $40 billion by FY2029 — the mobile-chip giant is now formally challenging Nvidia on its home turf.
What exactly is Dragonfly?
Dragonfly is Qualcomm's new brand for data-center silicon, spanning four product lines: AI accelerators, data-center CPUs, custom chips, and connectivity chips.
This means → Qualcomm is not testing the water with a single chip. It is fielding a full product matrix aimed squarely at Nvidia's data-center stack.
The core technology is HBC — high-bandwidth compute, an architecture that places processing units right next to memory. Qualcomm claims 6× the bandwidth per watt versus conventional HBM (high-bandwidth memory, the standard in today's Nvidia AI racks).
In plain terms = traditional designs keep the chip and memory far apart, wasting power shuttling data back and forth. HBC puts them side by side — 6× more data moved for the same energy bill.
Who is actually buying?
Meta is Qualcomm's first publicly named data-center customer, signing a multi-year CPU supply agreement for the C1000 and successor products.
CEOs of both Microsoft and Meta appeared in video messages, committing to be early adopters of Qualcomm's HBC and CPU products.
Qualcomm also disclosed custom-chip wins from "two major hyperscalers" and a deal with Saudi AI firm Humain to deploy 200 MW of Qualcomm accelerator racks starting this year.
This means → Qualcomm's customer list is no longer a promissory note. Meta has signed on paper, Microsoft has endorsed publicly — real contracts before the pitch deck, a different credibility level than a launch-day slide.
How aggressive is the C1000 CPU?
Dragonfly C1000 uses a chiplet design with 250+ cores, clock speeds above 5 GHz, support for PCIe Gen7, CXL, and LPDDR memory, with optional HBC.
Volume production is set for H2 2028; the first HBC chip ships with the AI250 rack in FY2027.
In plain terms = 250+ cores at 5 GHz+ would be top-tier specs for a data-center CPU. But in chips, "strong on paper" and "strong in production" are separated by at least a year of validation.
Why is the revenue target jumping so dramatically?
Data-center products are expected to contribute roughly $300 million this fiscal year; the FY2027 target is $5 billion — a 16×-plus increase in just over a year.
Total non-handset revenue guidance was raised from $22 billion to $40 billion by FY2029, with data-center accounting for about $15 billion.
The disclosure drove Qualcomm shares up ~15% after hours, reversing a 3%+ drop during regular trading.
This means → the market gave instant approval to Qualcomm's "second growth curve beyond phones," but leaping from $300 million to $5 billion carries extreme execution risk — investors are pricing this with an options-trade mentality.
How does China fit into the picture?
China contributed 46% of Qualcomm's revenue in FY2025 — the single largest market.
CEO Cristiano Amon said Qualcomm will design data-center chips for Chinese customers within U.S. export-control rules.
Bloomberg reported that Qualcomm has reached a deal with ByteDance to supply custom AI data-center chips structured to comply with current export-control thresholds.
This reflects Qualcomm's core dilemma: the data center is the new growth engine, but the biggest revenue source is China — any tightening of controls would squeeze both the old and new businesses simultaneously.
Can Nvidia's moat actually be breached?
Qualcomm announced an all-stock acquisition of chip-software startup Modular Inc. for roughly $4 billion, targeting Nvidia's CUDA software ecosystem — a toolkit that lets developers write AI programs on Nvidia GPUs, currently the industry's de facto standard.
Nvidia CEO Jensen Huang responded in kind: Nvidia systems deliver "the lowest cost per token, the highest token throughput, and the highest revenue," and the new Vera CPU is expected to generate nearly $20 billion in revenue this year.
BofA analyst Vivek Arya noted Qualcomm's current data-center revenue is "almost negligible" and called the AI market "hyper-competitive with large incumbents." BofA raised its price target from $165 to $195 but kept an underperform rating.
In plain terms = Wall Street's verdict is "direction approved, execution unproven" — they raised the target price but left the rating unchanged, which translates to "great story, show us the numbers first."
Content is for reference only, not financial advice.